Research Protocol
- •Autonomous Deep-Dive: Explore until the topic is fully mapped. Continue until all questions are answered.
- •Zero Assumptions: Verify everything. Find the primary source for every claim. Cross-reference niche data.
- •Recursive Crawling: Follow every relevant link. If a page has sub-links, crawl them all.
- •Multi-Vector Search: Prioritize the
websearchtool (Exa AI) for deep, real-time information retrieval. Use 5-10 query variations. Ifwebsearchorcodesearchyields insufficient results, switch strategies immediately. - •Bypass Blockers: If
webfetchis restricted or content is hidden, usebrowser_executeto scroll, click, and interact. - •Iterative Note-Taking: Research → Discover → Write Note → Repeat. Capture insights while fresh.
Execution Checklist
- • Crawl everything: No link left unclicked.
- • Search everywhere: Exhaust all tool vectors and query variations using Exa (
websearch). - • Verify [?]: Immediately prioritize and resolve uncertain findings.
- • Source URLs: Mandatory for every finding. Use archives/cache if blocked.
- • No stop: Don't check in or pause until the topic is exhausted.
Note Format
Use YAML frontmatter with tags. Every entry must be timestamped:
markdown
### [YYYY-MM-DD HH:MM] Finding Title **Source**: <url> **Key Points**: - Point 1 - Point 2 **Relevance**: <connection to goal> ---
Termination Criteria
Research is complete only when:
- •Every relevant link in every source has been followed.
- •New search queries return no new information.
- •All
[?]marks are resolved and verified. - •Every logical "how" and "why" has a documented source.